Exploring Key Point Analysis with Pairwise Generation and Graph Partitioning

TL;DR

The paper tackles the Key Point Analysis problem by addressing limitations of prior two-stage, intra-cluster-focused methods that rely on semantic similarity. It introduces a pairwise generation model that both detects whether two arguments share a key point and generates that key point, and then constructs a weighted argument graph whose edges encode the presence and content of shared key points. An iterative local-search graph partitioning algorithm operates on this graph to produce subgraphs representing distinct key points, with soft partition allowing arguments to participate in multiple subgraphs. Experiments on ArgKP and QAM show state-of-the-art performance, validating the benefits of combining pairwise generation with graph-based partitioning for robust, concise key-point extraction.

Abstract

Key Point Analysis (KPA), the summarization of multiple arguments into a concise collection of key points, continues to be a significant and unresolved issue within the field of argument mining. Existing models adapt a two-stage pipeline of clustering arguments or generating key points for argument clusters. This approach rely on semantic similarity instead of measuring the existence of shared key points among arguments. Additionally, it only models the intra-cluster relationship among arguments, disregarding the inter-cluster relationship between arguments that do not share key points. To address these limitations, we propose a novel approach for KPA with pairwise generation and graph partitioning. Our objective is to train a generative model that can simultaneously provide a score indicating the presence of shared key point between a pair of arguments and generate the shared key point. Subsequently, to map generated redundant key points to a concise set of key points, we proceed to construct an arguments graph by considering the arguments as vertices, the generated key points as edges, and the scores as edge weights. We then propose a graph partitioning algorithm to partition all arguments sharing the same key points to the same subgraph. Notably, our experimental findings demonstrate that our proposed model surpasses previous models when evaluated on both the ArgKP and QAM datasets.

Figures (4)

• Figure 1: An example in the ArgKP dataset. Only a subset of arguments and key points are shown due to space limitation. Our goal is to generate key points (right side) within a specific topic based on the provided arguments (left side). A single argument may correspond to multiple key points.
• Figure 2: Overview of our approach. Each argument possesses a stance representing its polarity ("pro" or "con"). For the sake of brevity, it has been omitted in the figure.
• Figure 3: Shared key point detection on ArgKP.
• Figure 4: Comparison of different models on key point generation on ArgKP. All the models were based on Flan-T5-large.